FEDERAL RESERVE BANK OF SAN FRANCISCO
WORKING PAPER SERIES
Working Paper 2007-06 http://www.frbsf.org/publications/economics/papers/2006/wp07-06bk.pdf
The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Federal Reserve Bank of San Francisco or the Board of Governors of the Federal Reserve System. This paper was produced under the auspices for the Center for the Study of Innovation and Productivity within the Economic Research Department of the Federal Reserve Bank of San Francisco.
Productivity Shocks in a Model with Vintage
Capital and Heterogeneous Labor
Milton H. Marquis Florida State University
and
Bharat Trehan
Federal Reserve Bank of San Francisco
January 2007
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGECAPITAL AND HETEROGENEOUS LABOR
MILTON H. MARQUIS AND BHARAT TREHAN
Abstract. We construct a vintage capital model in which worker skills lie
along a continuum and workers can be paired with different vintages (as tech-
nology evolves) under a matching rule of “best worker with the best machine”.
Labor reallocation in response to technology shocks has two key implications
for the wage premium. First, it limits both the magnitude and duration of
change in the wage premium following a (permanent) embodied technology
shock, so empirically plausible shocks do not lead to the kind of increases
in the wage premium observed in the U.S. during the 1980s and early 1990s
(though an increase in labor force heterogeneity does). Second, positive dis-
embodied technology shocks tend to push up the wage premium as well, and
while this effect is small, it does mean that a higher premium does not provide
unambiguous information about the underlying shock.
Labor reallocation also means that if embodied technology comes to play
a larger role in long-run growth, investment and savings tend to fall in steady
state, with little effect on output and employment, enabling the household
to increase consumption without sacrificing leisure. The short run effects are
more conventional: permanent shocks to disembodied technology induce a
strong wealth effect that reduces savings and induces a consumption boom
while permanent shocks to embodied technology induce dominant substitution
effects and an expansion characterized by an investment boom.
Date: January 29, 2007.
Key words and phrases. Vintage capital, wage premium, heterogeneous labor, productivity.
We thank Andrew McCallum and Jason Tjosvold for research assistance on this project. The
paper has benefitted from the feedback we received at seminars at Florida State University and
the Federal Reserve Bank of San Francisco. The opinions expressed in the paper are those of the
authors and do not necessarily reflect the views of the Federal Reserve Bank of San Francisco or
the Federal Reserve System.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR1
1. Introduction
In this paper we examine the effects of technology shocks in a model
with vintage capital and a heterogeneous labor force. We focus upon
two issues that have been discussed in the literature recently. One of
these has to do with the relationship between the wage premium paid
to high skilled workers and the growth rate of productivity. We ask,
for instance, whether empirically plausible shocks to the productivity of
capital can lead to increases in the wage premium that are comparable
to those seen in the data. The other issue concerns the economy’s
response to different kinds of technology shocks, that is, we ask whether
the economy reacts in the same way to embodied technology shocks
(which change the productivity of the latest vintage of capital relative
to earlier vintages) as it does to disembodied technology shocks (which
effect all vintages equally).
Positive embodied technology shocks lead to an increase in the wage
premium in our model, since the best workers are paired with the best
machines. However, this effect is small and not very long lived, because
of the key assumption that workers can be reallocated across vintages
of capital when the technology shock hits. Thus, firms respond to an
increase in the productivity of the latest vintage by reallocating labor
to the latest vintage and away from the older vintages. This tends
to depress the marginal product of labor (and hence the wage rate)
on the newest vintage and raise the marginal product of labor on the
oldest vintage, thereby mitigating the increase in the wage premium
that would otherwise occur
Much previous research has argued that embodied technology shocks
are behind the recent increase in the wage premium.1 In general, these
1See Katz and Murphy (1992) for an early example of the argument that increasing wage
dispersion is related to skill-biased technical change. Also see Autor, Katz and Krueger (1998),
Greenwood and Yorukoglu (1997) and Hornstein and Krusell (2002).
2 MILTON H. MARQUIS AND BHARAT TREHAN
models restrict the substitutability between skilled and unskilled la-
bor.2 For instance, Krusell, Ohanian, Rios-Rull, and Violante (KORV)
(1999) argue that the economy’s ever-improving technology requires
an ever more highly skilled workforce, and this pushes up the wages of
high-skilled workers relative to the low skilled. To us, the KORV story
seems at odds with the observation that new technology ultimately
permeates the entire workplace, eventually to benefit all workers.3 To
take a simple example, most people in the economy today can take a
high-quality photograph and operate a computer, which was not the
case when those technologies were first introduced. Moreover, the skill
required to perform those tasks is no more difficult to acquire than
performing the comparable tasks prior to the invention of the new
technology. Indeed, it is probably less difficult to master a modern-day
digital camera than to draw representations of reality on a sketch pad.
Utilizing new technologies may require a new and different set of skills,
but these skills are not necessarily more difficult to master than the
skills required to operate older technologies.4
Recent empirical evidence has not been entirely favorable to the hy-
pothesis that the rising wage premium is related to embodied produc-
tivity growth, either. For instance, Figure 1 shows that while the wage
premium rose quite sharply during the 1980s, it did not increase by
2Jovanovic(1998) is one exception.3This argument is similar to Aghion’s (2002) description of how a General Purpose Technology
spreads through the workforce and the economy.4From a modeling perspective, this distinction is not trivial. The characterization of the
evolution of the distribution of human capital in KORV’s model is a nonstationary process that
eventually must converge to a point mass of highly skilled workers in order to prevent the wage
premium from becoming infinite. The alternative characterization that we adopt in this paper is
a stationary distribution of human capital for which the wage premium remains bounded despite
continual advances in new technology that are embodied into the capital stock. Note that we are
not arguing that the mean skills of the workforce do not improve over time, as postulated for
example in the Lucas (1988)-Uzawa (1965) models of endogenous growth. We are arguing that it
makes little sense to require that the distribution of skills become degenerate asymptotically.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR3
nearly as much during the second half of the 1990s, which is a pe-
riod of unusually high productivity growth. This high productivity
growth is generally attributed to the IT revolution, which is probably
the most significant example of embodied technical change in the last
few decades. Similarly, the 1960s are a period of high productivity
growth as well, but Figure 1 shows that the wage premium did not
increase very much over this period either.
[FIG 1: LOG OF THE 90-10 WAGE PREMIUM]
The fact that the wage premium did not rise in the 1960s could
reflect a difference in the kind of productivity shocks that hit the econ-
omy at that time. It could be argued, for instance, that while embodied
technology shocks raise the wage premium, disembodied shocks do not.
However, our model provides evidence against this intuition. It turns
out that firms respond to a disembodied technology shock by real-
locating labor away from the newest vintage and towards the oldest
vintage. This has the effect of raising the marginal product of labor
on the newest vintage relative to the other vintages. Thus, the wage
premium goes up in response to disembodied technology shocks as well.
And while the parameterization of our model suggests that the increase
in the wage premium due to disembodied technology shocks is not as
large as that due to embodied technology shocks (of the same size), a
key implication is that an increase in the wage premium (even when
accompanied by an increase in productivity) cannot be used to argue
that the economy has been hit by an embodied productivity shock. The
model suggests that such a determination can only be made by looking
at what has happened to the amount of labor employed in different
jobs.
In any case, productivity shocks do not lead to changes in the wage
premium that are anywhere near the magnitude observed in the data,
suggesting that other factors (such as changes in the skill distribution
4 MILTON H. MARQUIS AND BHARAT TREHAN
of labor) are likely to be behind the observed changes in the skill pre-
mium.5 Consequently, in the final section of our paper we pose the
question of whether an exogenous increase in the variation of worker
skills, or a shift toward a more heterogeneous workforce that maintains
the same stock of human capital in the economy, could account for the
observed sizable increase in wage dispersion that occurred in the United
States during the 1980s and 1990s. We find that a calibration of the
model can easily be found which is consistent with that change. This
increase in wage dispersion has no effect on consumption, investment,
and output.
Our second issue concerns differences in the way that different kinds
of technology shocks affect key macroeconomic variables. Gilchrist and
Williams (GW) (2000) and Benhabib and Hobijn (BH) (2001) have
shown that vintage capital models (with undifferentiated labor) are
better able to reproduce important dynamic relationships in the econ-
omy than the standard RBC model. These include “hump-shaped”
responses of employment and output to technology shocks, and, as em-
phasized by GW, a weaker initial effect of the shock on employment
followed by a more drawn out return to the long-run equilibrium. BH
focus on the possibility that an acceleration in the rate of technological
advance could induce an investment boom in the economy analogous to
what appears to have characterized the U.S. economy in the late 1990s.
This result relies on an increase in the savings rate resulting from an
increase in the pace of embodied technological progress, which sets the
economy off on a permanently higher growth path for investment and
output.
We find that many of the predictions concerning the short-run dy-
namics of the vintage capital models of GW and BH described above
carry over to our model, though these dynamics become more com-
plex. In the short run, the savings rate rises in response to permanent
shocks to embodied technology, while it falls in response to permanent
5See Card and DiNardo (2002) for arguments against assigning technology the major role inthe recent rise in the wage premium.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR5
shocks to disembodied technology. These different responses are due
to the fact that the disembodied technology shock is realized across all
vintages of capital and induces a wealth response on the part of house-
holds, whereby they wish to increase their consumption at the expense
of investment, creating a consumption boom until the persistence in the
shock dies out. In contrast, a permanent shock to embodied technol-
ogy carries with it a transitory component that temporarily steepens
the quality gradient across vintages of capital until the new technology
has worked its way across all vintages. The households respond in the
short run with a dominant substitution effect that reduces consump-
tion’s share of output, thus increasing the savings rate, and bringing
about an investment boom. This response dies out as the persistence
in the shock subsides.
However, unlike BH, we find that the steady-state savings rate declines
as more of the economy’s productivity growth comes from embodied
rather than disembodied technology. The reallocation of workers across
machines plays a role here as well; households take advantage of the
steeper quality gradient that a more rapid development of embodied
technology affords by shifting the allocation of workers away from the
older technology and toward the newer technology, thus enhancing the
efficacy of the economy’s human capital in production. This allows con-
sumption to rise without requiring higher levels of output, and hence
a greater work effort.
2. Theoretical Model
The model makes use of the Lucas (1980) “multimember household”
abstraction to capture the heterogeneity of the workforce, where there
is a distribution of workers across a skill continuum of measure one.
The workers offer their labor services to an aggregate firm that op-
erates multiple production processes, each with a different vintage of
capital. The firm assigns worker groups to production processes ac-
cording to a matching rule that preserves an ordering of workers with
6 MILTON H. MARQUIS AND BHARAT TREHAN
the highest-skilled group of workers assigned to the latest vintage of
capital, the next highest-skilled group assigned to the second newest
vintage of capital, etc. As Jovanovic(1998) points out “It is plausi-
ble and empirically well founded to suppose that new technologies and
skills are complements, and so the new machines will be used by the
most skilled workers.” We further assume that the workers in each
group who are assigned to work in the same production process (i.e.,
with the same vintage of capital) receive the same wage. The wage
rate is based on the average labor quality of the group as a whole.
Growth in the economy results from both improvements in disem-
bodied technology, which affects all production processes symmetri-
cally, and embodied technological progress, which differentially affects
productivity across the production processes by determining the qual-
ity of the latest vintage. We assume that the number of vintages in the
economy is fixed, and enforced by an exogenous scrappage rule under
which the price of the oldest vintage of capital when it is sold for scrap
is determined. This price is taken to be a one-period projection from
the price of the oldest vintage based on the schedule of equilibrium
prices of capital stocks for which there are markets. This schedule of
prices declines with vintage.
2.1 Production with Vintage Capital and the Determinants of Growth
The production technology that we employ is linearly additive across
production processes that themselves are Cobb-Douglas in capital and
effective units of labor.6 The capital employed by each production
process is of a separate vintage, denoted j = 1...T , with the stock of
the latest vintage of capital in production at date t designated K1t and
the oldest vintage designated KTt .
6There is a literature on “jelly capital” in which an aggregation of capital across vintages is
possible when complementarities may be present across the differentiated production processes
that employ capital of a different vintage. See, e.g., Phelps (1962) and Benhabib and Hobijn(2001). The linearly additive production process that we utilize is a particular parameterization
of those models in which the elasticity of substitution across production processes is one.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR7
F(Kj
t , Hjt ; µt, At−j
)= µt
T∑j=1
(Kj
t
)α(At−jH
jt
)1−α
, α ∈ (0, 1), j = 1...T
(1)
In this expression, each vintage has a level of embodied technology at-
tached to it. This is captured by the term At−j and contributes to the
productivity of labor by scaling up each unit of human capital, Hjt ,
assigned to work with that vintage of capital. Human capital incorpo-
rates a quality adjustment to hours worked, as described in the next
subsection. Therefore, there are two enhancements to hours worked
assigned to any particular production process: one coming from em-
bodied technological progress and the other from the average skill level
of that particular group of workers, whose composition is endogenously
determined.
The two sources of long-run productivity growth are embodied tech-
nological change, γt, or the gross growth rate of At, where the current
history of these technological improvements is given by the sequence:
γt−j =( At−j
At−j−1
), j = 1...T (2)
and the gross growth rate of disembodied technology, gt, where:
gt =( µt
µt−1
)(3)
To work with a stationary model, we employ the following useful
normalization. Define the normalized capital stocks to be:
Kjt =
( Kjt
Ωt−j
), j = 1...T, (4)
8 MILTON H. MARQUIS AND BHARAT TREHAN
where the latest vintage of capital is normalized on the current level
of disembodied technology and the level of embodied technology asso-
ciated with the oldest vintage of capital used in current production,
or:
Ωt−1 = µ1
1−α
t At−T (5)
and more generally for any given vintage j :
Ωt−j = µ1
1−α
t At−T−j+1 (6)
The normalized production function is given by:
F(Kj
t , Hjt ; Gt−i, Γt−j
)=
T∑j=1
( Kjt∏j−2
i=0 Gt−i
)α(Γj
tHjt
)1−α
, α ∈ (0, 1), j = 1...T
(7)
where we define:
Gt−i =(Ωt−i−1
Ωt−i−2
)(8)
(with∏−1
i=0 Gt−i = 1). The level of embodied technology in the jth
vintage of capital relative to the level of embodied technology in the
oldest (T th) vintage is given by:
Γjt =
( At−j
At−T
), j = 1...T (9)
Note that:
Γ1t > Γ2
t > ... > ΓTt = 1, (10)
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR9
which fully captures the “quality gradient” in capital across vintages.
The more rapid is the pace of embodied technological progress, the
steeper is this “quality gradient.”
As will become apparent below, under this normalization, the equi-
librium gross growth rate of the economy at date t is simply given by
Gt, and can be expressed as the following combination of disembodied
and embodied technological progress:
Gt = g1
1−α
t γt−T (11)
2.2 Workforce Heterogeneity
The multimember household has a continuum of P workers indexed
by x, and distributed uniformly along the unit interval: x ∈ [0, 1]. The
skill level of a worker with index x is denoted by the function h(x),
which determines the quality-adjustment to the units of labor supplied
by that worker such that the most highly skilled workers have an index
value of x = 0 and a skill level of h(0), while the least-skilled workers
have an index value of x = 1 and a skill level of h(1).
It is assumed that the labor-leisure decision is collectively taken by
the household, with all workers choosing the same fraction of leisure
time, denoted lt, and labor, denoted zt. The household’s total leisure
time for all workers is given by Lt = Plt. With the time allocation
decision for each worker given by zt + lt ≤ 1, the total amount of time
allocated to labor is:
Pzt = P − Lt (12)
Given a discrete (and constant) number of vintages of capital, de-
noted T , in use at any point in time, the quality-adjusted units of labor,
or human capital, that are allocated to vintage j are given by:
10 MILTON H. MARQUIS AND BHARAT TREHAN
Hjt = Pzt
∫ xjt
xj−1t
h(x)dx
= Pztχ(x)∣∣∣xj
t
xj−1t
= Pzt
[χ(xj
t)− χ(xj−1t )
], j = 1, ..., T (13)
where ztχ(x) is the cumulative distribution of human capital per capita
employed at date t. The total number of hours worked by workers
assigned to capital of vintage j is given by:
N jt = Pzt
∫ xjt
xj−1t
dx = Pzt
[xj
t − xj−1t
], j = 1...T (14)
2.3 Household Optimization
A schedule of prices of the various vintages of capital goods is re-
quired to establish the scrappage rule. Therefore, we choose to place
the ownership of the capital goods with the household, and employ a
Lucas (1978)-style asset pricing formulation. We also have chosen to
have the household select contingent group employment decision rules
by partitioning x into T worker groups of contiguous skill levels, de-
noted xjt = xj
t − xj−1t , j = 1...T that are being offered to the firm.
Given the skill profile of the household’s workers within each group,
these choices are constrained by the “matching rule” (of “best work-
ers” with the “best machines”) and determined in equilibrium by the
firm’s demand for units of quality-adjusted labor, Hjt , j = 1...T . The
household also makes its consumption/savings and labor/leisure deci-
sions.
The household’s optimization problem is:
maxct,Lt,x
jt ,Hj
t ,Kjt+1
E0
∞∑t=0
βU(ct, Lt), j = 1...T β ∈ (0, 1) (15)
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR11
where the household’s consumption is ct and its capital holdings are
Kjt T
j=1, whose values are given at date t=0 when the optimization is
conducted.
Each period, the household faces a budget constraint:
ct+K1t+1+
T−1∑j=1
pjtK
j+1t+1 ≤
T∑t=0
RjtK
jt +
T∑t=0
W jt Hj
t +T∑
t=1
pjt(1−δ)Kj
t , δ ∈ (0, 1)
(16)
under which the household combines: its capital income, where Rjt is
the rental rate on a unit of the jth vintage of capital; its labor income,
where W jt is the wage rate per unit of quality-adjusted labor (or human
capital) assigned to the jth production process; and its revenue from
the sale of its capital holdings, where pjt is the market price of the jth
vintage of capital at date t and δ is the depreciation rate, in order to:
make its consumption purchases; complete its new investment in the
latest vintage of capital, K1t+1, which it will rent to the firm beginning
in period t + 1; and acquire its holdings of used capital to be carried
over to next period.
The household’s labor supply decisions are further constrained by its
total available human capital and the matching rule, such that:
Hjt = Pzt
[χ(xi
t
)− χ
(xi−1
t
)], j = 1, ..., T (17)
and
T∑j=1
xjt =
T∑j=1
[xi
t − xi−1t
]= 1 (18)
It also faces the time resource constraint:
zt +Lt
P≤ 1 (19)
12 MILTON H. MARQUIS AND BHARAT TREHAN
After normalizing the household’s problem on Ωt−j, and defining the
normalized variables ct =(
ct
Ωt−1
)and wj
t =(
W jt
Ωt−1
), the Euler equations
become:
βE(Uct+1
Uct
)[Rj+1t+1 + pj+1
t+1(1− δ)
Gt+1pjt
]= 1, p0
t = 1, j = 0...(T − 1)
(20)
wjt = wj−1
t , j = 1...T (21)
Uct
T∑j=1
Hjt w
jt = ztPULt (22)
Equation (20) is the collection of familiar asset pricing equations for
those vintages of capital for which there is a market. This excludes
vintage T , which is scrapped at the end of the period. Equation (21)
equates the wage rate paid per unit of human capital (quality-adjusted
labor) across all worker groups. (Hourly wages and the wage premium
are discussed below.) Equation (22) is the optimal labor/leisure deci-
sion that holds for all workers.
2.4 Firm’s Optimization
The firm is assumed to be competitive in the factor and product
markets. It rents physical capital of all vintages from households and
hires quality-adjusted units of labor that are sorted into worker groups
and matched with capital vintages by quality. Each member of a given
group works in the same production process and receives the same
wage. The firm performs the following static period-by-period profit-
maximization:
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR13
maxKj
t ,Hjt
[µt
T∑j=1
(Kjt )
α(At−jHjt )
(1−α) −T∑
i=1
RjtK
jt −
T∑j=1
W jt Hj
t
], j = 1...T
(23)
where period profits are expressed as output less the factor payments.
After normalizing on Ωt−j, the firm’s first-order conditions yield a
set of expressions where each factor payment is equated to the factor’s
marginal product.
α(Kj
t
Hjt
)α−1( j−2∏i=0
Gt−i
)1−α(Γj
t
)1−α
= Rjt , j = 1...T (24)
(1− α)(Kj
t
Hjt
)α( j−2∏i=0
Gt−i
)−α(Γj
t
)1−α
= wjt , j = 1...T (25)
Note that the normalized wage rates on quality-adjusted labor, or hu-
man capital, wjt , tend to be positively affected by embodied technology,
as indicated by Γjt , the quality gradient on vintage capital. The im-
plication is that the firm will wish to increase the labor allocation to
later vintages that possess a higher level of embodied technology. This
effect becomes more pronounced as the quality gradient steepens.
The “hourly” wages paid to each worker can be determined by first
computing the equilibrium normalized wage bill paid by the firm to
each worker group. For the production process employing vintage j
capital, the normalized wage bill is wjtH
jt , where Hj
t equals the to-
tal quality-adjusted units of labor employed. The average normal-
ized hourly wage rate for workers assigned to vintage j capital is
vjt = wj
tHjt /N
jt . The corresponding hourly wage rate is vj
t = Ωt−1vjt .
Note that, whereas a steepening of the quality gradient of capital across
vintages tends to increase the wage differentials across worker groups, a
14 MILTON H. MARQUIS AND BHARAT TREHAN
reallocation of workers toward the latest vintages has the effect of low-
ering the average quality of workers in every worker group, and thus
tending to lower wages economy-wide.
One measure of the “wage premium” paid to workers assigned to
more recent vintages of capital is the ratio of the wage rate per worker
for that (j) vintage to the wage rate per worker for the least-skilled
workers utilizing the oldest vintage, or vjt /v
Tt . The wage premium paid
to the most highly skilled workers relative to the least skilled is v1t /v
Tt .
A steeper quality gradient across capital vintages will tend to raise
the wage premium, while a reallocation of workers toward the latest
vintages will tend to reduce the wage premium.
2.5 Equilibrium
Equilibrium in the goods market consists of transforming output
goods along with the scrapped capital of vintage T , the sum of which
is defined as normalized output, yt, into consumption and new invest-
ment. After normalizing,
ct + it = yt =T∑
j=1
( Kjt∏j−2
i=0 Gt−i
)α(Γj
tHjt
)1−α
+ pTt (1− δ)
( KTt∏T−2
i=0 Gt−i
)(26)
where normalized investment is given by: it = Gt+1K1t+1, and the un-
depreciated portion of the oldest vintage of capital is sold at the end
of the period for a unit price of pTt as determined by the exogenous
scrappage rule7:
7Gilchrist and Williams (2000) endogenize the scrappage decision in a vintage capital model.However, they achieve endogenous scrappage by relying on an ex post Leontieff technology, witha “one-man-one-machine” constraint that does not allow for an optimal ex post reallocation of
workers across vintages to be characterized by endogenous capital-labor ratios. This modelingchoice is unsuitable for our purposes, since an optimal deployment of a heterogeneous workforceacross vintages of capital is essential to the issues addressed in this paper.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR15
pTt
pT−1t
=pT−1
t
pT−2t
(27)
The evolution of the normalized capital stocks is given by:
Kj+1t+1 = (1− δ)Kj
t (28)
with an associated movement down the quality gradient, from Γjt−1 to
Γj+1t , reflecting the aging of the technology embodied in the capital
stock.
3. Calibration Issues
To calibrate the model, choices must be made for the utility func-
tion, U(c, L) and the distribution of skills, h(x). The utility function,
expressed in terms of normalized consumption is assumed to be log-
linear:
U(ct, Lt; Ωt−1) = ln(Ωt−1ct
)+ η ln Lt, η > 0 (29)
For the distribution of skills, an exponential function was found to fit
well the empirical distribution of human capital estimated by Abowd,
Lengermann, and McKinney (2002). Their estimates are based on the
Longitudinal Employer-Household Dynamics (LEHD) Program data
for 1992.8 The empirical model is:
h(x) = S0eφ(1−x), S0, φ > 0 (30)
8These data cover California, Illinois, Michigan, and North Carolina for the 1st quarter, andinclude over 400,000 observations. See Abowd, Lengermann, and McKinney (2002), Table 9.
16 MILTON H. MARQUIS AND BHARAT TREHAN
and the estimated values were: S0 = 8.92 and φ = 2.187. 9 Figure 2
displays the values from Abowd, et. al. together with the fitted values
from the regression.
[FIG 2: DISTRIBUTION OF HUMAN CAPITAL]
3.1 Steady-State Model
The model is solved for a small set of vintages, T = 3, which is suffi-
cient to highlight the basic properties of the model. In the steady-state,
normalized version of the model, there are: 21 endogenous variables,
Hj, xj, L, z, pj, rj, c, Kj, wj, j = 1...3, where the net real rental rates
on capital Kj, j = 1...3 are given by rj = Rj − δ, j = 1...3; 4 exoge-
nous variables Γj, G, j = 1...3; and 7 parameters, A, φ, β, δ, η, α, P .
Therefore, 11 constraints are needed to solve the model. We make the
following selections. S0 and φ are estimated as described above. Capi-
tal’s share of income is set to α = 0.33. Having limited the number of
vintages to T = 3, the depreciation rate is set to δ = 1/T .10 The popu-
lation of workers is an exogenous scale variable in the model that is set
to P = 100. The annualized gross growth rate of the economy is set to
G = 1.025, of which 60 percent is attributed to embodied technologi-
cal progress, or Γj = 1.015, j = 1, 2, 3. The contribution of embodied
technology to growth is consistent with Greenwood, Hercowitz, and
Krusell (1997), who relied on Gordon’s (1990) relative price series for
producer durables versus consumer nondurables and services, and with
Gilchrist and Williams who estimated the contribution of embodied
technology in a combined “putty-clay”-Solow vintage capital model to
be between 50 and 70 percent, and it is similar to the 2/3 estimate
9The estimated regression was Lhc = 9.21 + 0.0219Pe, where Lhc denotes the log of the dollar
value of human capital and Pe denotes the percentile of the human capital distribution. Theequation had an adjusted-R2 of 0.98.
10This assumption yields very high rental rates on capital, but avoids excessively high scrap-
page values. With extraordinary complications to the model, T could be increased sufficiently tobring δ down to reasonable levels. However, this should not change the qualitative properties of
the model that we wish to highlight.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR17
of Wilson (2000) based on estimates of production functions for the
manufacturing sector.
The remaining parameters, β = 0.9847, and η = 1.7437, were chosen
to be consistent with a 40-hour workweek (z = 0.36) and a real net
rental rate on the most recent vintage of capital of 6.08 percent, which is
broadly consistent with the long-run return on U.S. equities [see Mehra
and Prescott (1985)]. As a check on our calibration, we computed the
equivalent of the 90-10 wage premium in our model to be v1
v3 = 1.408
and plotted this number (which is based on data for 1992) on the graph
displaying the actual 90-10 wage premium in Figure 1. The model
generated value turns out to be quite close to the data.
Table 1 reports a summary of the parameters and steady-state values
derived from this calibration exercise that we use as the “benchmark”
in subsequent exercises.
[INSERT TABLE 1: Benchmark Parameters and Steady-State
Values]
To solve the stochastic version of the model and run dynamic sim-
ulations, characterizations of the stochastic processes driving the two
sources of productivity growth are needed. We assumed ARIMA(1,1,0)
processes for each, reflecting the fact that a time series on average pro-
ductivity is well characterized by such a process. However, little guid-
ance is available on how the variance of productivity growth should be
decomposed between disembodied technology shocks (µ) and embodied
technological progress (A). For the purposes of this paper, we assumed
the processes to be symmetric, apart from their respective trends.
From equation (11), the stochastic process for Gt is thus derived
from the processes for µt and At as follows:
ln gt = (1−ρm) ln g +ρm ln gt−1 + εt, ρm ∈ [0, 1), εt ∼ N(0, σ2m) (31)
18 MILTON H. MARQUIS AND BHARAT TREHAN
ln γt = (1−ρA) ln γ +ρA ln γt−1 +νt, ρA ∈ [0, 1), νt ∼ N(0, σ2A) (32)
where ε is the growth rate shock to disembodied technology, and ν is
the growth rate shock to embodied technology. Then, from equations
(11), (31), and (32):
ln Gt+1 = ln(Ωt+1
Ωt
)= (1− α)−1 ln gt+1 + ln γt−T+1 (33)
and the mean gross growth rate for the economy, G, is given by:
G = g1
1−α γ (34)
The stochastic model was estimated using the undetermined coef-
ficients method described in Christiano (2002). For the simulation
exercises, we set σA = (1 − α)−1σm = 0.020, which approximates the
percent standard deviation of the annual growth rate of output for the
U.S. economy over the period 1963-2002 (where the sample has been
selected to match the wage data). For the persistence parameters, we
searched over the range of values ρA, ρm ∈ [0, 1), and computed the
selected second moments from the data for each pair based on a sam-
ple of trial runs involving 5000 replications of length 39 (to match the
sample period). The results are broadly consistent with those reported
in Gilchrist and Williams (2000) and are not reported in the paper,
but are available from the authors on request.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR19
4. The Effects of Embodied versus DisembodiedTechnical Change
This section reports on exercises designed to examine two issues con-
cerning the nature of productivity growth. The first set of exercises
focuses on the consequences of shifting the overall composition of the
sources of long-run steady-state growth from the benchmark settings
of 60 percent embodied technology/40 percent disembodied technol-
ogy to 100 percent embodied technology. This shift in the sources
of productivity growth results in a steepening of the quality gradient
across vintages of capital and induces a reallocation of labor toward the
more recent vintage with very little change in employment. The “level
effects” that the economy experiences include a decline in aggregate
output that is accompanied by a significant reduction in the savings
rate that enables aggregate consumption to rise.
The second set of exercises compares the dynamic responses of the
economy to the two sets of shocks. The wage premium goes up after the
embodied technology shock, but this increase is temporary and does
not appear to be large enough to explain the increase observed in the
U.S. Somewhat surprisingly, the wage premium goes up after a disem-
bodied technology shock as well. By contrast, the responses of output,
investment and consumption to the two shocks turn out to be quali-
tatively different, due to the transitory aspect of (permanent) shocks
to embodied technology that is absent in the shocks to disembodied
technology.
4.1 Steady State Effects of Changes in the Source of Growth
Using the benchmark parameter settings, and maintaining a long-
run growth rate of 2.5 percent (G = 1.025), the long-run balanced
growth properties of the model are computed for an economy in which
growth is assumed to come entirely from embodied technical progress
(that is, an economy where γ = 1.025). The results are compared to
the benchmark values in Table 2.
20 MILTON H. MARQUIS AND BHARAT TREHAN
[INSERT TABLE 2, STEADY-STATE COMPARISONS WITH
100% EMBODIED TECH GROWTH]
Along the new long-run growth path, neither the normalized level of
consumption nor output are affected very significantly, with the former
rising by four-tenths of a percent while the latter falls by three tenths
of a percent. This discrepancy is attributable to investment, which
declines by about 1-1/2 percent. Consistent with this decline, the sav-
ings rate falls by slightly more than 1 percent as well. Note that this
is in contrast to the findings of Benhabib and Hobijn (2001) who find
that (in a model with vintage capital and a homogenous labor force)
both the savings rate and investment rise when the pace of embodied
technical progress goes up. In our model, a steepening of the quality
gradient of capital leads to a reallocation of labor towards the newest
vintage, which allows for a more efficient use of the economy’s heteroge-
nous labor stock. (Table 2 shows that the employment share allocated
to the newest vintage rises by nearly 2 percent while that allocated to
the oldest vintage falls by more than 1-1/2 percent.) In equilibrium,
households increase consumption somewhat and allow investment to
fall by more.
As expected, the steeper capital quality gradient is also reflected in
a change in the relative rental rates on the different vintages of capital,
with the rental rate on the newest vintage rising by more than 1 per-
centage point while the rental on the oldest vintage falls slightly. Per-
haps more surprising is the finding that all three wage rates–including
the wage rate for workers on the newest vintage– in the model decrease.
Two offsetting forces are at work on the wages of workers assigned to
the newest vintage. By itself, the steeper capital-quality gradient tends
to push up the wages of these workers. However, the workers that are
newly assigned to the latest vintage have lower human capital than
the workers already working on these machines, and this tends to re-
duce the average wages for this group. Hourly wages fall for workers
on other vintages as well, reflecting the reallocation of the relatively
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR21
higher skilled workers towards the newer vintages. The net effect is a
small increase in the wage premium.
4.2 Comparing the Response of the Economy to Permanent Shocks to
Embodied versus Disembodied Technology
Growth rate shocks to embodied and disembodied technology are as-
sumed to contribute equally to the variance of output in the benchmark
model. The processes are assumed to be symmetric, with persistence
parameters of 0.2, which is the estimate one obtains when real output
growth is regressed on its own lag over our 40 year sample period.
The qualitatively different responses of the savings rate to the two
permanent shocks are illustrated in the top panel of Figure 3, which
displays the impulse response functions resulting from one-standard
deviation shocks to embodied (solid lines) and disembodied technology
(dashed lines). A permanent (positive) shock to embodied technology
leads to an increase in the savings rate during the first two periods, as
this is the period during which an investment in new capital with im-
proved technology can temporarily exploit the steeper vintage-capital
quality gradient. The increase in investment can be seen in the mid-
dle panel of Figure 3. Investment falls off because the steeper quality
gradient eventually dissipates once the shock has reached the oldest
vintage. The lowest panel of the figure shows that part of the increase
in investment following an embodied technology shock is achieved by
a drop in consumption during the first two periods. In contrast, the
permanent disembodied technology shock induces an increase in con-
sumption (middle panel of figure 3) which is accommodated partly by
a reduction in the savings rate and investment.
This figure illustrates a key difference in the response of the economy
to these shocks: A permanent shock to embodied technology elicits an
investment boom, with the savings rate rising while a permanent shock
to disembodied technology is followed by a consumption boom, with
a declining savings rate. These differential responses owe to the fact
22 MILTON H. MARQUIS AND BHARAT TREHAN
that, unlike a permanent shock to disembodied technology, a perma-
nent shock to embodied technology has a transitory component as-
sociated with it, as the improved technology works its way through
the vintages of capital over time. This transitory feature of the shock
strengthens the substitution effect relative to the wealth effect and pro-
vides a greater incentive to invest. The disembodied technology shock
has no such temporary character, as it simultaneously affects all pro-
duction processes in a similar manner. The optimal response of the
household is to increase consumption, as the wealth effect dominates
the substitution effect in the consumption-savings decision.
[INSERT FIGURE 3: RESPONSE OF SAVINGS, INVESTMENT,
CONSUMPTION]
[INSERT FIGURE 4: RESPONSE OF OUTPUT AND EMPLOY-
MENT]
Figure 4 illustrates the output and employment responses to perma-
nent embodied and disembodied technology shocks. As the top panel
indicates, both kinds of shocks tend to push up output, though the ef-
fect of embodied technology shocks is larger and more persistent. The
same thing is true in the labor market. While the initial response of
total labor hours to an embodied technology shock is similar in mag-
nitude to the disembodied technology shock, the embodied technology
shock has a more persistent effect. This would suggest that strength in
the labor market should be evident for a more extended period of time
following an investment boom than would would observe following a
consumption boom.11
[INSERT FIGURE 5: LABOR ALLOCATIONS AND WAGE PRE-
MIUM]
11It appears that the persistence in the strong labor market response would be enhanced inthis model if the number of vintages of capital were increased.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR23
Figure 5 shows what happens to the allocations of labor (hours)
across the different vintages of capital and to the wage premium in
response to the two kinds of technology shocks. In response to an
embodied technology shock, labor gets allocated toward the newest
vintage (panel A) and away from the oldest capital vintage (panel C).
This response is pronounced for periods 2 and 3 when the results of
capital investment decisions associated with the improved technology
are differentially affecting the vintages of capital to the greatest extent.
The response to a permanent disembodied technology shock tends to
be more muted in terms of the skill allocations across vintages, but the
direction of reallocation is the opposite of that which takes place in
response to embodied technology shocks.
As expected, both shocks tend to push up wages. The permanent
nature of the productivity shocks, whether to embodied or disembodied
technology, induces an upward shift of the growth path for wages that
benefits all workers. With the size of the shocks identical, the long-
run effects on wages are identical. There is a difference in the short
run response, though. Following an embodied technology shock, the
premium paid to workers utilizing the most recent vintage of capital
relative to those assigned to the oldest vintage increases, with a spike in
that measure of the wage premium evident in period 3 (see panel D of
Figure 5). Period 3 is the last period for which the embodied technology
has not influenced all vintages. It is the period when workers who are
assigned to the oldest vintages are at the greatest disadvantage and
this disadvantage is reflected in their productivity, and hence in their
wages. The wage premium tends to go up after an embodied technology
shock as well, though the increase is noticeably smaller.
[INSERT FIGURE 6: LABOR AND WAGES RESPONSES TO A
SEQUENCE OF SHOCKS]
To highlight the differences in labor re-allocation in response to the
two kinds of shocks, Figure 6 shows what would happen if the economy
24 MILTON H. MARQUIS AND BHARAT TREHAN
were to be hit by a sequence of five shocks, similar, perhaps to what
might have happened during the late 1990s. Embodied technology
shocks lead to a pronounced reallocation of labor away from the oldest
vintage, while the labor allocated to both the newest and the middle
vintage goes up. The wage premium goes up as well. By contrast,
labor gets reallocated in the reverse direction following a disembodied
technology shock. This reallocation seems surprising at first glance,
since the disembodied technology shock affects all vintages equally.
However, the fact that all vintages are affected equally means that the
disembodied technology shock changes the productivity of the newest
vintage relative to the oldest vintage, and so causes a reallocation of
labor.
The disembodied technology shock also causes an increase in the
wage premium. Though this increase is not as large as that which re-
sults from an embodied technology shock, it does have the interesting
implication that a higher wage premium –even when accompanied by
rising productivity– can not be interpreted as conclusive evidence of
embodied technical change. In our model, the two shocks are distin-
guished by what happens to employment across the different vintages,
which suggests that one should look at employment across skill levels
to determine the nature of the underlying shock.
5. Workforce Heterogeneity and the Wage Premium
While the evidence in the previous section suggests that faster tech-
nological progress will tend to push up the wage premium, the increases
we obtain are neither very large nor very long lived. This suggests that
labor demand factors that are technology-driven are not likely to pro-
duce the very large increases in wage dispersion that was observed in
the U.S. economy during the 1980s and 1990s.12
12As discussed above, KORV argue that complementarities between skilled labor and producerdurables in production may be capable of explaining the wage premium between 1960 and 1985.
We do not explore this possibility in our model.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR25
This section looks to the supply side of the labor market for an alter-
native mechanism that could induce large changes in wage dispersion.
It turns out that the introduction of greater heterogeneity into the dis-
tribution of skills across workers, while the aggregate stock of human
capital in the economy is held fixed, leads to a significant increase in
wage dispersion, with very little or no consequence for the macroecon-
omy.
In the experiment that we conduct, all parameter values are fixed
at their benchmark settings, except those that determine the distri-
bution of human capital, ie., S0 and φ. These two parameters are
adjusted to match as nearly as possible the average 90-10 wage pre-
mium for the U.S. economy over the 1963-1980 period, a time when
wage premium (3.2 by our measure) was noticeably lower than the
benchmark period of 1992 (when the estimated premium is 4.1), while
maintaining the benchmark value for the aggregate stock of human
capital (H = 1161.2). The resulting distribution of human capital (la-
belled “low-heterogeneity distribution”) is displayed against the actual
data and the benchmark distribution from the model (labelled “fitted”)
in Figure 2. As can be seen, the low-heterogeneity distribution is flat-
ter than the original; it turns out that the two are also statistically
distinguishable. Table 3 compares the resulting steady state (labelled
“less heterogenous human capital”) with the steady state under the
benchmark specification.
With a more homogenous workforce, the benchmark allocation of hu-
man capital becomes inefficient. To restore equality in the wages paid
per unit of human capital (wj′s), a more symmetric allocation of labor
is required. Thus labor is reallocated away from the oldest vintage and
toward the latest vintage. As reported in Table 3, this reallocation is
substantial, with labor allocated to the oldest vintage of capital falling
sharply (i.e., by nearly 12 percent), while the allocation to the newest
vintage rises substantially (i.e., by more than 13 percent); there is also
a modest increase in the labor allocated to the second oldest vintage
26 MILTON H. MARQUIS AND BHARAT TREHAN
(which goes up by roughly 5 percent). This redeployment of the work-
force that results from the flattening of the economy’s human capital
distribution is accompanied by a reduction in the wage premium. The
highest hourly wage (v1) falls by nearly 12 percent while the lowest
hourly wage (v3) rises by more than 13 percent.
Due to the constant returns to scale technology, with a unitary elas-
ticity of substitution of human capital between production processes,
the change in the degree of heterogeneity of the workforce has almost
no effect on consumption, investment, or output. These results reflect
the unrealistic assumption that the stock of human capital remains un-
changed as the degree of heterogeneity of worker skills changes. Even
so, it allows us to illustrate how labor supply effects can dramatically af-
fect the wage premium, in contrast to the exercises reported in the pre-
vious section which suggested that technology-driven demand for high
skilled workers is unlikely to be sufficiently strong to induce changes in
the wage premium that are comparable to those observed in the U.S.
economy in the 1980s and 1990s.
6. Conclusions
This paper examines ways in which heterogeneity in the skill levels
of the workforce may matter for an economy that employs a variety of
technologies in production, and where the mix of those technologies is
constantly evolving. When the most highly skilled workers are matched
with the latest technologies, a reallocation of workers is always under-
way that aims to improve macroeconomic performance. Our theoretical
results suggest that when embodied technological progress accounts for
an increasing share of long-run economic growth, the greater efficacy
of the workforce that this reallocation affords induces a decline in the
savings rate with little change in output and employment, thus per-
mitting an increase in aggregate consumption without the sacrifice of
leisure. In the short-run, permanent shocks to embodied technology
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR27
have a transitory component to them because it takes time to “ab-
sorb” the new technology across vintages. This transitory feature of
the shocks induces dominant substitution effects that raise the savings
rate, increase employment, and stimulate output that is characterized
by a investment boom. In contrast, permanent shocks to TFP induce
a strong wealth effect in the consumption-savings decision that lowers
the savings rate. In this case, employment and output rise, but the
decomposition of output is indicative of a consumption boom.
Labor reallocation also plays an important role in how the wage pre-
mium behaves after a technology shock. In the case of an embodied
technology shock, it tends to limit both the duration and the magnitude
of the increase in the wage premium that follows an embodied technol-
ogy shock. Somewhat surprisingly, labor reallocation also means that
the wage premium goes up following a disembodied technology shock,
because the economy responds by moving labor away from the newest
and towards the oldest vintage.
In any case, neither of these two effects appears to be strong enough
to account for the increases in wage dispersion that have been observed
in the U.S. economy during the 1980s and 1990s. Our conclusion is that
one must look to the supply side of the labor market for answers to this
puzzle. Efforts to understand how the growth of human capital in the
economy has been accompanied by distributional effects (with respect
to worker skills) would likely be a fruitful avenue of future research.
28 MILTON H. MARQUIS AND BHARAT TREHAN
Table 1: Benchmark Parameters and Steady-State Values
Parameters Benchmark Variables Steady-State Variables Steady-State
Values Values Values
α 0.33 c 840.81 z 0.36
β 0.9847 i 520.37 x1 0.26
δ 1/3 y 1361.18 x2 0.31
η 1.7437 K1 507.68 x3 0.43
γ 1.015 K2 338.45 N1 9.47
G 1.025 K3 225.63 N2 11.11
P 100 r1 6.080∗ N3 15.42
T 3 r2 4.906∗ H1 642.35
S0 10.01 r3 3.768∗ H2 405.35
φ 2.187 p1 0.970 H3 255.79
ρA = ρm 0.2 p2 0.941 v1 42.90
σA 0.021 p3 0.913 v2 23.08
σm 0.03 v3 10.49
* in percent
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR29
Table 2: Steady-State Effects of Changes in Embodied
Technology’s Share of Growth
variable/ benchmark All growth from embodied tech.
statistic γ = 1.015 γ = 1.025
percentage change
normalized consumption, c 840.81 0.4
normalized investment, i 520.37 -1.4
normalized output, y 1361.18 -0.3
emp. share allocated to vintage 1, x1 0.26 1.9
emp. share allocated to vintage 2, x2 0.31 0.7
emp. share allocated to vintage 3, x3 0.43 -1.6
total hours worked, N1 + N2 + N3 36.00 0.04
real net rental rate on K1, r1 6.08 1.3*
real net rental rate on K2, r2 4.90 0.5*
real net rental rate on K3, r3 3.77 -0.3*
normalized hourly wage: vintage 1, v1 42.90 -0.1
normalized hourly wage: vintage 2, v2 23.08 -0.9
normalized hourly wage: vintage 3, v3 10.49 -0.5
savings rate,(
iy
)38.23 -1.1
log of the wage premium, log(
v1
v3
)1.41 0.3**
variance of log wages 3.76 0.1
* changes in the percent rates of return
** percentage change of the ratio(
v1
v3
)
30 MILTON H. MARQUIS AND BHARAT TREHAN
Table 3: Steady-State Effects of Changes in Labor Force
Heterogeneity
variable/ benchmark low-heterogeneity case
statistic (percentage change)
labor hours allocated to vintage 1, N1 9.470 13.3
labor hours allocated to vintage 2, N2 11.11 5.1
labor hours allocated to vintage 2, N3 15.42 -11.9
total stock of human capital, H1 + H2 + H3 1303.49 no change
normalized hourly wage: vintage 1, v1 42.90 -11.8
normalized hourly wage: vintage 2, v2 23.08 -4.9
normalized hourly wage: vintage 3, v3 10.49 13.5
wage premium,(
v1
v3
)4.09 -22.2
log of the wage premium 1.41 -17.8
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR31
References
Abowd, John M., Paul A. Lengermann, and Kevin J. McKinney. (2002)
“The Measurement of Human Capital in the U.S. Economy”, U.S. Cen-
sus Bureau LEHD Program Technical paper No. TP-2002-09.
Aghion, Phillipe. (2002)“Schumpeterian Growth Theory and the Dy-
namics of Income Inequality.” Econometrica, May, pp. 855-882.
Autor, David, Lawrence Katz and Alan Krueger. (1998)“Computing
Inequality: Have Computers Changed the Labor Market?” Quarterly
Journal of Economics, November, pp. 1169-1214.
Benhabib, Jess and Bart Hobijn. (2001) Another View of Investment:
40 Years Later, in: J.Stiglitz, R.Frydman, M.Woodford, and P.Aghion
(eds.), Knowledge, Information and Expectations in Modern Macro-
economics: In Honor of Edmund S. Phelps, 2003, 522- 545, Princeton
University Press: Princeton
Card, David and John DiNardo. (2002)“Skill Biased Technological
Change and Rising Inequality: Some Problems and Puzzles.” Journal
of Labor Economics, 20(4), pp. 733-783.
Christiano, Lawrence J.(2002) “Solving Dynamic Equilibrium Models
by a Method of Undetermined Coefficients.” Computational Economics
20(1-2) pp. 21-55
Gilchrist, Simon and John Williams. (2000) ”Putty-Clay and Invest-
ment: A Business Cycle Analysis.” Journal of Political Economy, vol-
ume 108, pages 928960.
Gordon, Robert. (1990) The Measurement of Durable Goods Prices,
Chicago: University of Chicago Press.
32 MILTON H. MARQUIS AND BHARAT TREHAN
Greenwood, Jeremy and Mehmet Yorukoglu. (1997)”1974”, Carnegie-
Rochester Conference Series on Public Policy, v. 46, pp. 49-95.
Greenwood, Jeremy, Zvi Hercowitz, and Per Krusell. (1997) ”Long-
Run Implications of Investment-Specific Technological Change”, Amer-
ican Economic Review, 87:3, pp. 342- 362.
Hornstein, Andreas and Per Krusell. (2002)“From factor-specific tech-
nical change to wage inequality and back again: A quantitative-theoretic
study of the macroeconomic effects of IT”, Federal Reserve Bank of
Richmond Economic Quarterly, pp. 1-28.
Jovanovic, Boyan. (1998) “Vintage Capital and Inequality”, Review of
Economic Dynamics, vol. 1, pp. 497-530.
Katz, Lawrence and Kevin Murphy. (1992)“Changes in Relative Wages,
1963-1987: Supply and Demand Factors.” Quarterly Journal of Eco-
nomics, February, pp. 33-60.
Krusell, Per, Lee Ohanian, Jose-Victor Rios-Rull and Giovanni Vi-
olante. (2000) “Capital-Skill Complementarity and Inequality: A Macro-
economic Analysis”, Econometrica, September, pp. 1029-1053.
Lucas, Robert E., Jr. (1980) ”Equilibrium in a Pure Currency Econ-
omy,” Economic Inquiry, vol 18(2), pp.203-220.
Phelps, Edmund S. (1962) “The New View of Investment: A Neoclas-
sical Analysis”, Quarterly Journal of Economics, 78(1), pp. 172-176.
Mehra, Rajnish, and Edward C. Prescott. (1985) “The equity pre-
mium: a puzzle,” Journal of Monetary Economics, 15, pp. 145-161.
PRODUCTIVITY SHOCKS IN A MODEL WITH VINTAGE CAPITAL AND HETEROGENEOUS LABOR33
Wilson, Dan. (2001) ”Capital-Embodied Technical Change: Mea-
surement and Productivity Effects.” Ph.D. Dissertation, University of
Maryland, May.
Florida State University
E-mail address: [email protected]
Federal Reserve Bank of San Francisco
E-mail address: [email protected]
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 1000
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
FittedActual
Percentile
Dollars
'Low Heterogeneity'distribution
1960 1965 1970 1975 1980 1985 1990 1995 20001
1.1
1.2
1.3
1.4
1.5
1.6
Model generated value
Fig 1. Log(90-10) Wage Premium
Fig 2. Human Capital – Actual and Fitted Values
34
Fig 3. Responses to a Single Technology Shock - I
0 3 6 9 12 150.374
0.378
0.382
0.386
0.390
0.394
Embodied shock
Disembodied shock
Years
A. Savings Rate
0 3 6 9 12 15455
460
465
470
475
480
Embodied shock
Disembodied shock
Years
B. Investment
0 3 6 9 12 15740
745
750
755
760
765
770
Embodied shock
Disembodied shock
Years
C. Consumption
35
Fig 4. Responses to a Single Technology Shock - II
0 3 6 9 12 151205
1210
1215
1220
1225
1230
1235
Embodied shock
Disembodied shock
Years
A.Output
0 3 6 9 12 1535.9
36
36.1
36.2
36.3
36.4
Embodied shock
Disembodied shock
Years
B. Labor Hours
36
Fig
5. R
espo
nses
to a
Sin
gle
Tech
nolo
gy S
hock
-III
03
69
1215
0.26
0
0.26
2
0.26
4
0.26
6
0.26
8
0.27
0
0.27
2
Em
bodi
ed s
hock
Dis
embo
died
sho
ck
Yea
rs
A. L
abor
Allo
cate
d to
New
est V
inta
ge
03
69
1215
0.30
6
0.30
8
0.31
0
0.31
2
0.31
4
0.31
6
Em
bodi
ed s
hock
Dis
embo
died
sho
ck
Yea
rs
B. L
abor
Allo
cate
d to
Mid
dle
Vint
age
03
69
1215
0.41
4
0.41
8
0.42
2
0.42
6
0.43
0
0.43
4
Em
bodi
ed s
hock
Dis
embo
died
sho
ck Yea
rs
C. L
abor
Allo
cate
d to
Old
est V
inta
ge
03
69
1215
4.06
4.08
4.10
4.12
4.14
4.16
4.18
Em
bodi
ed s
hock
Dis
embo
died
sho
ck
Yea
rs
D. W
age
Prem
ium
37
Fig
6. T
he E
ffect
of a
Seq
uenc
e of
Tec
hnol
ogy
Sho
cks
03
69
1215
4.06
4.08
4.10
4.12
4.14
4.16
4.18
4.20
4.22
4.24
Em
bodi
ed s
hock
Dis
embo
died
sho
ck
Yea
rs
D. W
age
Prem
ium
03
69
1215
0.25
8
0.26
2
0.26
6
0.27
0
0.27
4
0.27
8
0.28
2
Em
bodi
ed s
hock
Dis
embo
died
sho
ck Yea
rs
A. L
abor
Allo
cate
d to
New
est V
inta
ge
03
69
1215
0.30
4
0.30
6
0.30
8
0.31
0
0.31
2
0.31
4
0.31
6
0.31
8
Em
bodi
ed s
hock
Dis
embo
died
sho
ck
Yea
rs
B. L
abor
Allo
cate
d to
Mid
dle
Vint
age
03
69
1215
0.40
0.41
0.42
0.43
0.44
Em
bodi
ed s
hock
Dis
embo
died
sho
ck
Yea
rs
C. L
abor
Allo
cate
d to
Old
est V
inta
ge
38